Discussion Intel current and future Lakes & Rapids thread

[ondma]

Intel went with 4 cores for the U lineup for a number of reasons. The decision was made before AMD even released an 8-core part.

Also, the H series is 8 cores and launches this week.

Yea, but will the chips actually be available?

 

[LP-ZX100C]

https://twitter.com/x/status/1391424722257080322

You must add personal commentary to your post in the CPU forum.
No dropping of links or video without comment.


esquared
Anandtech Forum Director

 

[jpiniero]

Yea, but will the chips actually be available?

Remains to be seen but you'll have plenty of models to choose from at least, it looks.

https://twitter.com/x/status/1391424722257080322

Sounds about right for Intel. Anything 7 nm in 2023 almost has to be tiny chiplets and/or very low volume.

 

[eek2121]

<snip>

Psst, posts like this aren't allowed

 

[eek2121]

Remains to be seen but you'll have plenty of models to choose from at least, it looks.



Sounds about right for Intel. Anything 7 nm in 2023 almost has to be tiny chiplets and/or very low volume.

TSMC is taking over the world.

 

[clemsyn]

TSMC is taking over the world.

And if China is successful in taking over Tiawan, then China is taking over the world.

 

[DrMrLordX]

@clemsyn

shush we get scolded for bringing in politics. Needless to say, TSMC can start moving operations elsewhere/building fabs elsewhere if they feel the need. Samsung has already started testing the waters (desert?) in the United States.

@LP-ZX100C

@eek2121 is right that posts without commentary are generally not allowed, but given the content (which is solid gold) I think people can mostly overlook that. The Intel capex and projected future capex is stunning. 2-3% on EUV equipment? There's no way they're going to fab anything in significant quantity on 7nm or beyond at any point in the future with capex spending like that. Stick a fork in em, Intel is done. Assuming that data is accurate.

 

[Gideon]

This twitter thread is interesting:

https://twitter.com/x/status/1391613641808506880

Things to note as far as Intel's EUV goes
1. ASML is completely booked till 2024. Intel literally can't just order more within that timeframe.
2. Intel cancelled some of their EUV orders last year, which Samsung and TSMC used to order up a lot of them. (Worrying, but it was done before Gelsinger became CEO)
3. The further down you go in terms of process node, the more EUV you need. By 2023, Intel will be on 7nm, roughly equivalent to TSMC 4.3nm. TSMC and Samsung, on the other hand will be on risk production of 2nm and high volume production of 3nm.

 

[JasonLD]

Well, TSMC needs EUVs for 7nm+ and all smaller process nodes, and Samsung for all 7nm and smaller.
Intel only needs EUVs for their 7nm and pretty much the only 7nm chips(or rather chiplets) they will be manufacturing will be compute tiles for Meteor Lake based processors in 2023.

 

[JoeRambo]

Intel only needs EUVs for their 7nm and pretty much the only 7nm chips(or rather chiplets) they will be manufacturing will be compute tiles for Meteor Lake based processors in 2023.

And the amount of EUV depends on actual process "recipe". Not all layers need EUV and ironically Intel has the most experience in industry with ridiculous amount of SAQP done. 6 years of ramping up bad yields?
So short term they are probably fine from "machine count" number, the real question is if 7nm is healthy at all, that is where confidence of everyone including Intel seems to end currently.

 

[dacostafilipe]

So short term they are probably fine from "machine count" number, the real question is if 7nm is healthy at all, that is where confidence of everyone including Intel seems to end currently.

Well, even if 7nm is a big success, without the EUV machines to gear up all the FABs, capacity will be an issue.

 

[DrMrLordX]

Well, even if 7nm is a big success, without the EUV machines to gear up all the FABs, capacity will be an issue.

Exactly. If you look at the projection data from chia kokhua/Mizuho, you will see that they project 20kwpm by 2023 for Intel, which is ridiculously small, and seems like a rosy projection.

 

[Gideon]

Exactly. If you look at the projection data from chia kokhua/Mizuho, you will see that they project 20kwpm by 2023 for Intel, which is ridiculously small, and seems like a rosy projection.

And that's the year when both Samsung and TSMC are (at least in risk production) with GAAFETs. Intel obviously isn't going anywhere with the absurd manufacturing deficit we have, but I also don't see them getting back to competition with TSMC @ the leading edge. If anything they might even fall behind Samsung.

EDIT: in fact, the very same slide projects that TSMC is doing 20kwpm for GAAFET 2nm by 2023.

 

[exquisitechar]

Exactly. If you look at the projection data from chia kokhua/Mizuho, you will see that they project 20kwpm by 2023 for Intel, which is ridiculously small, and seems like a rosy projection.

I'd take that with a pinch of salt, some of their data doesn't look accurate. Makes me question their estimates for the future too.

 

[DrMrLordX]

I'd take that with a pinch of salt, some of their data doesn't look accurate. Makes me question their estimates for the future too.

Well it is the second report I've seen indicating that Intel hasn't been buying much EUV equipment. The exact wpm numbers may be off, but how will we ever know?

 

[Dayman1225]

Well it is the second report I've seen indicating that Intel hasn't been buying much EUV equipment. The exact wpm numbers may be off, but how will we ever know?

We will know when we see 7nm ramp.

 

[jpiniero]

We will know when we see 7nm ramp.

Doesn't seem like that will really be in 2023. Could just be 24 or 25.

 

[Gideon]

So Alder Lake seems to be a November product:

Intel Core "Alder Lake-S" to Launch in November 2021

Intel's 11th Gen Core "Rocket Lake-S" could have a brief stint as the tip of Intel's desktop spear. Wccftech reports that its successor, the 12th Gen Core "Alder Lake-S," could launch by November 2021. The processor debuts on the new LGA1700 socket, and will introduce many firsts to the...

www.techpowerup.com

 

[Ajay]

Do we know how many EUV masks Intel is using on it's 7nm?

 

[Ajay]

This twitter thread is interesting:

https://twitter.com/x/status/1391613641808506880

This prediction is interesting, but it is old. It predicts the launch of Intel 7nm in 2H19! Maybe the first chart is more recent than the second one? TSMC CAPEX looks about right (depending on CAPEX includes R&D, which hasn't been clearly established yet).

 

[eek2121]

So Alder Lake seems to be a November product:

Intel Core "Alder Lake-S" to Launch in November 2021

Intel's 11th Gen Core "Rocket Lake-S" could have a brief stint as the tip of Intel's desktop spear. Wccftech reports that its successor, the 12th Gen Core "Alder Lake-S," could launch by November 2021. The processor debuts on the new LGA1700 socket, and will introduce many firsts to the...

www.techpowerup.com

The article mentions it is on 10SF, isn’t ADL-S on ESF? I imagine we won’t see enthusiast chips until January.

 

[Hitman928]

This prediction is interesting, but it is old. It predicts the launch of Intel 7nm in 2H19! Maybe the first chart is more recent than the second one? TSMC CAPEX looks about right (depending on CAPEX includes R&D, which hasn't been clearly established yet).

I don't think the chart is saying Intel 7 nm launched in 2H19. They are listing that as risk production with 0 capacity. I'm assuming they are using the term risk production as a broad term. If you look at the length of the 7 nm risk production and the estimated capacities, it looks like they are predicting a similar path of Intel's 10 nm where they can technically produce wafers, but the process is not ready for actual production yield and won't really be ready for any type of volume before 2023 (that's where the chart ends at least).

 

[Ajay]

I don't think the chart is saying Intel 7 nm launched in 2H19. They are listing that as risk production with 0 capacity. I'm assuming they are using the term risk production as a broad term. If you look at the length of the 7 nm risk production and the estimated capacities, it looks like they are predicting a similar path of Intel's 10 nm where they can technically produce wafers, but the process is not ready for actual production yield and won't really be ready for any type of volume before 2023 (that's where the chart ends at least).

Okay, I see the labeling I obviously missed before. Duh. 20K WPM is a pittance for Intel, it really isn't HVM. Intel must have cancelled more than 2 AMSL EUV machines (perhaps they had 'options' on additional devices which they chose not to act on).

 

[Hitman928]

Okay, I see the labeling I obviously missed before. Duh. 20 WPM is a pittance for Intel, it really isn't HVM. Intel must have cancelled more than 2 AMSL EUV machines (perhaps they had 'options' on additional devices which they chose not to act on).

I don't know how accurate this market research report is, but if it's even close to accurate, it doesn't paint a very good picture for Intel's recovery but rather points to a lot more pain in their future.

 

[blckgrffn]

Okay, I see the labeling I obviously missed before. Duh. 20 WPM is a pittance for Intel, it really isn't HVM. Intel must have cancelled more than 2 AMSL EUV machines (perhaps they had 'options' on additional devices which they chose not to act on).

That really baffles me. I know they are expensive, but the finance guys love to depreciate stuff. And if they are on the super, duper critical path for future growth that seems hugely irresponsible to everyone, even the shareholders.

Wouldn't you buy them just so the competition didn't get them at some point? I am thinking about the (really old now) Only the Paranoid Survive and I feel like that Intel might have been busy burying knives into backs to keep their dominance.